A lightweight C library for creating, training, and evaluating simple neural networks with customizable initialization methods. This library supports basic forward and backward propagation, sigmoid activation, data handling, and model saving/loading.

---

Features

1. Custom Initialization Techniques: Supports zero, random, and Xavier initialization.
2. Forward and Backward Propagation: Implements core functions for feedforward and backpropagation.
3. Data Handling: Functions to initialize, normalize, partition, and describe data.
4. Model Serialization: Save and load model configurations to/from files.

Usage

1. Model Initialization: Set input, hidden, and output sizes with your preferred initialization method.
2. Training and Testing: Use train and test functions with Data and Parameters structures to train the model.
3. Model Persistence: Save and load models using save_model and load_model.

API Overview

Core Structures

1. Model: Represents the neural network with layer sizes and weight initialization.
2. Data: Holds data samples and labels for training/testing.
3. Parameters: Configurable training parameters.

Key Functions

1. Model Initialization: initialize_model()
2. Forward and Backward Propagation: forward(), backward()
3. Data Manipulation: normalize_data(), partition_data(), describe_data()
4. Model Persistence: save_model(), load_model()

Getting Started

git submodule update --init --recursive
bash build.sh

Example

#include "malpractice.h"

int main() {
    // Model parameters
    size_t input_size = 784, hidden_size = 128, output_size = 10;
    Model_InitTechnique init_tech = Model_Init_Xavier;
    Model *model = initialize_model(input_size, hidden_size, output_size, init_tech);

    // Load data (replace with actual loading logic)
    Data *data = zero_initialize_data(input_size, 100);

    // Set training parameters
    Parameters params = {.learning_rate = 0.01, .epochs = 1000, .log_train_metrics = 1};

    // Train and evaluate
    train(data, params, model);
    test(data, model);

    // Save model
    save_model(model, "model.bin");

    // Cleanup
    deinitialize_data(data);
    deinitialize_model(model);
    return 0;
}